Research On Group Target Tracking Based On Probability Hypothesis Density Filter

Target tracking is a basic problem in the real world.With the development of modern detection technology,the progress of weapons and equipment,and the increase of the complexity of target and environment,the target tracking system is facing many new problems and challenges.Tracking problems in dense multi-target scenes,including group target tracking,have attracted more and more attention.The group targets are composed of a series of targets with close spatial distance and similar motion characteristics.In this paper,according to the radar resolution,group target tracking is divided into two aspects: unresolvable group target tracking and resolvable group target tracking.Combined with the filtering algorithm under the framework of random finite set(RFS),these two aspects are studied respectively.The main work includes:(1)Research on unresolvable group target tracking algorithm.In order to solve the problem of indistinguishable targets in a group,it is considered to track the whole group directly.The Gauss mixture probability hypothesis density(GMPHD)group targets tracking algorithm based on elliptic random hypersurface model(RHM)is studied.Then,aiming at the problems of inaccurate measurement partition when group targets cross,large number of partition and large amount of computation in the above algorithms,a novel group target GMPHD filtering algorithm based on mean shift(MS)and bilayer group structure(BGS)model is proposed.By using the mean shift algorithm for measurement partition,the number of measurement partition is greatly reduced,and the operation amount is also greatly decreased.At the same time,a new BGS model is proposed to solve the problem of underestimation at the intersection time of group targets.By constructing the second layer group structure model,the relationship between different groups is explored,and the information is fed back to the measurement partition step to guide the measurement partition at the next time.The improved algorithm solves the problem of underestimation at the intersection time of group targets,reduces the number of measurement partition,and decreases the computational complexity of the algorithm.(2)Research on resolvable group target tracking algorithm.In this case that the targets in a group can be distinguished,if we continue to use unresolvable group target tracking algorithm mentioned above to track these targets,the estimation results might be too rough,and a lot of information about the group structure might be lost,and these information might play a crucial role in the subsequent judgment and decision.In order to solve this problem,we use the evolution network model to model for group structure,and combine with the box particle PHD filter to track the group target.The algorithm can dynamically obtain the state of each target in the group and group structure characteristics,and then realize the stable tracking for the whole group.On this basis,in order to improve the estimation performance of the algorithm,the cardinality distribution of the number of targets is introduced,and a novel box particle CPHD group target tracking algorithm based on evolution network model is proposed.Compared with the traditional particle CPHD group target tracking algorithm,the proposed algorithm greatly reduces the number of particles required for sampling and the amount of computation.And compared with the box particle PHD group target tracking algorithm,the proposed algorithm has better estimation performance in strong clutter environment and low detection probability environment.